Use of CFD in onshore facility explosion siting studies

SYMPOSIUM SERIES NO. 156 Hazards XXII # 2011 IChemE

USE OF CFD IN ONSHORE FACILITY EXPLOSION SITING STUDIES

Olav R. Hansen1, Scott Davis2 and Filippo Gavelli2

1GexCon AS, Bergen, Norway2GexCon US, Bethesda, Maryland, USA

Significant releases (on the order of 50–100 kg/s) of hydrocarbons, whether as flashing liquid

or dense gas, combined with moderate winds can, in less than one minute, generate very large

flammable vapour clouds in an onshore facility. This potential has been realized in several acci-

dents, both in the past (e.g., Flixborough, 1974), as well as more recently (e.g., Jaipur and San

Juan). In onshore siting studies for facilities, whether driven by the API RP-752 or the Seveso-

II/III directive, the typical approach is to use blast curve methods like the Multi-Energy method

(MEM), the Congestion Assessment Method (CAM2) or the Baker-Strehlow-Tang (BST). When

applying such methods, the typical approach is to only consider blast energy for the flammables

inside one unit of the plant at a time (one congested area). This assumption may be acceptable if

the spacing between units is sufficient and if there is no risk for deflagration-to-detonation-tran-

sition (DDT). However, recent accidents like the Buncefield explosion tell us that DDT cannot

be easily ruled out and that, if that assumption is made, the typical blast-curve approach may be

far from conservative. Another significant limitation of the blast-curve approach is that only few

mitigation solutions can be evaluated. In this paper a CFD-based approach is presented which

can be used to evaluate: i) the minimum (critical) gas cloud sizes to achieve DDT for different

areas of an onshore facility; ii) the potential for generation of large gas cloud sizes from the dis-

persion of gaseous or liquid releases; and iii) if required, the effectiveness of various mitigation

methods (e.g., soft barriers, confinement, deluge) to limit DDT-potential.

INTRODUCTIONRecently there have been several severe vapour cloudexplosion accidents in onshore facilities around the world.At the end of 2009, two massive explosions occurred attank farms in San Juan, Puerto Rico and Jaipur, India,with significant damage off-site. Both of these accidentshad similarities to the Buncefield explosion of 2005(BMIIB, 2009): in fact, in San Juan windows were reportedbroken 2–3 km away from the site, while 12 people losttheir lives in the Jaipur explosion. Significant explosionshave already occurred in 2010 as well, with multiple fatal-ities both at a petrochemical plant in Lanzhou, China, andthe Tesoro Refinery, Anacortes, USA.

According to international standards ISO 13702(1999) and ISO 19901-3 (2010) explosion risk studies onoffshore oil and gas installations are required to be per-formed using validated consequence models that are ableto take into account the effect of geometry confinementand congestion, and to evaluate mitigation options. Assuch, CFD tools are typically required by both of these stan-dards. Safety requirements are generally functional (i.e.,performance-based) rather than prescriptive. A risk studywill therefore have to evaluate how the consequences ofvapour cloud explosions may be kept below a certainthreshold through both design and mitigation, to minimizethe likelihood of collapse of structures or the failure ofbarriers.

Risk assessment studies for onshore facilities – atleast in countries adopting API-RP 752 or national regu-lations based on the Seveso-II directive – instead tend tobe simpler and driven by the “credible worst-case”approach. For a given plant, the largest units (congested

20

areas) are identified, and are assumed to be filled with astoichiometric flammable air/gas mixture. The explosionstrength of the flammable cloud is assumed, based on sub-jective congestion/confinement considerations, and there-after blast strength isopleths are calculated using a set ofenergy curves. Blast loads are then estimated for buildingsintended for occupancy, and if these structures meet theexplosion criteria, it satisfies the regulatory requirements.The blast loads, however, are growing with the flammablegas cloud volume assumed as “maximum-credible” (e.g.,smaller gas clouds give smaller blast loads).

One implicit assumption in the worst-case approachis that only the flammable cloud inside a congested regionof the plant will contribute to the blast energy. This assump-tion is based on the fact that deflagration flames, which aredriven by turbulence, will decelerate once the flames leavethe turbulence-generating congested region. Under theseconditions, the use of “safe” separation distances betweendifferent congested units may thus limit the energy fromeach independent congested region contributing to the blastwaves instead of grouping the regions and flammable cloudstogether.

There is however one major condition to the validityof this assumption: it is only valid in the deflagrationregime. If the flames accelerate to high enough flamespeeds (on the order of 1000 m/s), there is a potential riskfor deflagration-to-detonation transition (DDT). Detonationflames propagate by shock-ignition, i.e., shock-waves aheadof the flame auto-ignite to generate new shock-waves.Detonation flames therefore need no turbulence to sustain,and will continue to propagate at high velocity as long asthe gas concentration stays within the detonation limits


and the gas cloud thickness is above �13 detonation cellsizes (Desbordes, 1995). For gases like propane, this meansthat a 1.5 m thick flammable cloud may propagate a detona-tion, for methane the cloud needs to be more than 4 m thick.If a detonation occurs, it can often have a dramatic effecton the far field blast pressures. Not only can the blastenergy contributing to the shockwaves be one or twoorders of magnitude higher, but the blast “epicenter” mayalso get closer to the buildings of concern as flammablegas outside the “congested” areas may also contribute tothe explosion severity. As such, the efficacy of safety gapsor regions outside the congested area may be nullified ifDDT occurs.

The possibility of DDT or detonation is not cited inthe API-RP 752 (2009) standard and it is rarely consideredeven when risk studies are performed according to theSeveso-II directive. But are detonations that unlikely?

21

The Buncefield Major Incident Investigation Board(BMIIB, 2009) concluded in their final report that a tran-sition to detonation likely occurred during the Buncefieldpetroleum vapour explosion in 2005. The Buncefield siteconsisted mainly of large tanks and limited piping, and itwas initially unclear how the flames may have acceleratedto the point of a DDT. It was ultimately determined thatthe dense vegetation along the roads surrounding the site pro-vided the congestion necessary to accelerate the flames into alikely detonation. Vegetation within and around the plant isalso suspected to have provided the necessary congestionfor flame accelerations in the San Juan and Jaipur explosions,although DDT may not have occurred in those accidents.

In recent years, accidental detonations are likely notunique to the Buncefield incident. In August 2008, theSunrise Propane explosion occurred at an LPG facility inToronto (Ontario Fire Marshal 2010 report, see Figure 1).

Figure 1. Three frames from a CCTV camera observing Sunrise Propane explosion from a distance. The very fast flame acceleration

into the third picture (.1000 m/s), a very intense light, as well as videos captured from other angles seeing flames accelerate through

vegetation were among the reasons for concluding a DDT was seen (Ontario Fire Marshal, 2010). Courtesy of Ontatio Fire Marshals

Office.


The report indicated that during an LPG transfer from onetruck to another, a flashing release of LPG at 10 kg/s tookplace for possibly up to 15–20 minutes, creating a largeflammable propane cloud at the facility. FLACS CFD simu-lations used in the investigation demonstrated that the facil-ity could be covered by propane gas in around 3 minutes.CCTV cameras located a distance from the facility recordedthe explosion event, where one or two frames showed a verybright flame and unconfined (open area) flame speedsexceeding 1000 m/s. This event was concluded to likelybe a DDT. Two possible explanations for the DDT wereconsidered: it was caused by flames burning throughthousands of stacked propane bottles, or the more likelyexplanation: video footage indicated that the flames acceler-ated through trees near the site.

With this in mind, we should consider the followingquestions:

1. Are DDTs such unlikely events in onshore petro-chemical facilities that they should not be consideredin risk studies?

2. If DDTs cannot be ruled out, can the current API-RP752 and similar Seveso-II approaches be consideredacceptable?

LIMITATIONS WITH THE CURRENT SIMPLIFIED

APPROACH FOR ONSHORE RISK STUDIES?API-RP 752 mentions two typical approaches for choosingthe flammable gas cloud size to be used with blast curves:either to assume a filled congested volume or a smallerdispersion calculated congested volume.

The integral dispersion models, typically applied topredict the smaller dispersion calculated volume, are notcapable of resolving geometry effects. For instance: windspeeds inside a process unit will be much lower than thatmeasured outside the unit due to object congestion; andrecirculation zones/wakes due to partial confinement mayactually trap gas instead of allowing it to dissipate in thewind. These integral models often have limited capabilityfor low wind scenarios, where gravity and structures can

22

prevent dissipation of a flammable cloud, as was seen atBuncefield, or for release scenarios involving high momen-tum jet releases upwind, which are later blown back intothe facility in a very uniform concentration, possibly nearstoichiometric concentrations.

Detailed CFD simulation studies showed that thepotential to generate very large vapour clouds is significant,when evaluating sizeable releases of pressurized dense flam-mable gases, and in particular, flashing releases. For similarhole-sizes and operating pressures, a flashing release of aliquid may typically give release rates 3–4 times higher(material dependent) than the equivalent release of a gas.The flashing release also has a different mixing mechanismthan the gaseous release, resulting in lower velocities andvelocity gradients. A consequence of the phase transitionand energy balance, homogeneous vapour concentrations (ata low temperature) can occur at the distance where all theparticles have evaporated. For large flashing releases, the reac-tive cloud size can easily be 100 times larger than the con-gested volumes considered for the blast study (see examplein Figure 2). As the density is typically high for the flashingreleases, there will be limited vertical mixing outside the jetregion. Wind speeds above 2 m/s may however lift some ofthe dense cloud. One should also keep in mind that inventorysizes may be very large for tanks containing superheatedliquids (e.g., LPG), and it may be much more difficult todesign shut-down systems or pressure relief systems comparedto a pressurized gas system. A flashing release may thus emptya significant part of a tank content without any significantreduction in release rate (other than that caused by reducedhydrostatic pressure as the tank empties).

Flashing liquid substances with high boiling pointsreleased at elevated temperatures (30–508C above theboiling point) may evaporate more gradually and notfully evaporate, often resulting in a very homogeneousgas mixture with some additional liquid particles. If thishomogeneous gas concentration is between LFL and stoi-chiometry, this may give a large, very uniform, highlyexplosive vapour cloud in the event of an ignition, as thepassing flames may evaporate just enough flammable fog

Figure 2. Flashing releases 50–100 kg/s can within minutes fill large facilities with highly reactive flammable mixtures. The

illustration shows flammable cloud predicted with FLACS from a 400 LPG release (200 kg/s) during low wind conditions (2 m/s),

the leak direction is against the wind. Traditional consequence studies will tend to strongly underestimate the flammable gas

plume, as the integral dispersion models ignore geometry and can not handle low wind scenarios.


for optimal (stoichiometric) combustion (Hansen, 2004).The Flixborough accident seems to have been such a scen-ario (HSE, 1975).

As a conclusion to this part, the typical approachesused in onshore facilities for determining flammable gascloud sizes can underestimate the hazardous gas clouds byorders of magnitude. Also the use of integral models for dis-persion studies in onshore facilities will be very inaccurate,one should be particularly concerned about flashing liquidreleases when evaluating potential for the generation oflarge highly reactive gas clouds, due to potentially higherrelease rates and differences in the mixing mechanism.

The blast-curve approaches typically also haveseveral weaknesses, including limited ability to predict theactual explosion source strength or dynamics. Some models(Pierazio, 2005; Puttock, 1995) have developed relationsfor source strength based on experimental data, mostly atscales much smaller than plant-scale, and using subjective,averaged assumptions of congestion and confinement. Inparticular, relations for congestion level (volume blockageis a poor measure for flame acceleration potential) may bequestionable when scaled-up to real plant scale. If usedwith conservative assumptions regarding energy/cloud sizeand source pressure, however, the blast curve methodsmay efficiently provide reasonable and valuable far-fieldpressures. CFD tools can be used in combination withblast-curves like TNO-Multi-Energy Method (van denBerg, 1985), to predict both the source energy and generalpressure level.

“Conservative” assumptions should include the possi-bility for DDT and detonation of the entire vapour cloudoutside a congested unit, like discussed in the introduction.This is seldom done in API-RP 752 or Seveso-II studies, butcan change the scenario from a tolerable accident to somecompletely unacceptable.

One of the main problems with using too simplifiedconsequence tools and approaches for onshore explosionsiting studies is that the effect of geometry and most mitiga-tion methods cannot be predicted. This prevents innovativedevelopment of gradually safer design and improved miti-gation concepts. On the other hand, for offshore oil andgas installations, the more functional requirements andrisk criteria in standards like ISO 13702 (1999) and ISO19901-3 (2010), stimulate (and actually require) the evalu-ation of layout options and mitigation, forcing the industryto be innovative with regard to safety. Since the simple con-sequence tools are unable to accurately evaluate the effectof mitigation methods like change of layout and wall/deck configurations or water deluge, the plant operatorcannot know if a given mitigation measure has a net positiveor negative effect on the total risk, making it difficult tojustify investments in mitigation measures.

PROPOSED IMPROVEMENTS FOR ONSHORE

SITING STUDIESTo improve certain limitations with typical onshoreexplosion studies, an alternative CFD-based approach isproposed next. The details of the method have similarities

23

to approaches used on oil and gas offshore installations,and would provide recommendations to the recent additionsto API-RP 752, where advanced blast tools such as CFD arenow a recognized tool for risk-base approaches. There willbe significant focus on establishing accurate explosionpressures inside the plant, and how to mitigate explosionconsequences to prevent intolerable pressure levels andparticularly DDTs.

The study can either be carried out as a worst-caseapproach, identifying the worst possible consequences andhow to mitigate, or as a probabilistic risk assessment withthe goal to limit the probability of intolerable, escalatingscenarios to one every ten thousand years (1024) or someother tolerable limit.

Since this approach will be using CFD, there is a needfor a detailed description of the geometry layout. Differentoptions exist:

. For some of the modern facilities, reasonably accurateCAD-models may be available and can be directlyimported into the CFD consequence software.

. If no CAD-model exists, one option is to perform alaser-scan of the facility and generate a 3D model.This approach may be expensive, and can often have aprohibitive cost for a project.

. A third option is a manual implementation of a geometrymodel of the facility.

Manually modeling a complex facility in detail mayrequire several man-months and have a high associatedcost. As such, there are more efficient approaches such asthe representative congestion screening method, RCM,(Hansen et al., 2010, see Figure 3). RCM requires that themain structures, confining walls/decks and largest vesselsare manually implemented, and all the remaining conges-tion will be represented by repeated idealized obstructions.With this approach, a geometry model can be establishedwithin a few days, with the accuracy of the model increasingas the effort level increases. For all of the above approaches,there may be a need to evaluate that the congestion level indifferent parts of the plant is representative of what isexpected for this type of facility. Optimally, the congestionlevel would be checked by a site visit or review of photos orvideos of the site. If analysis of the geometry model (pipelengths for different pipe diameters or object surface area)indicate that the congestion is too low, it is common touse anticipated congestion methods to increase the conges-tion level.

The consequence study will consist of an explosionstudy and in most cases a dispersion study. Depending onthe approach chosen, the study can be performed in manydifferent ways. The three approaches include: (1) theworst-case approach; (2) the realistic worst-case approach;and (3) probabilistic risk analysis.

The worst-case approach will require performingnumerous explosion simulations for a range of largevapour clouds that are located in different parts of theplant. Different ignition locations are used to ensure that aDDT cannot occur even with very large, near stoichiometric


Figure 3. Accurate geometry model (left) and representative congestion model (RCM, right) are shown with the same area/volume

congestion (Hansen et al., 2010). Slightly higher (more conservative) pressures are predicted in explosion simulations using RCM (see

lower plots, unit is kPa).

vapour clouds spanning more than one congested processunit. If the possibility of DDTs can be ruled out, the pre-dicted explosion consequences can be used to generateworst-case blast contours (pressure or pressure impulse),including blast dynamic effects like reflections on buildingsand shielding by walls or process units.

The realistic worst-case approach will require simulat-ing numerous dispersion scenarios, by varying releaselocation, rate, direction, wind direction and speed, for bothgas and flashing liquid releases (if applicable), in orderto identify one or more potentially worst-case scenariosamong the range of release scenarios that may occur at thefacility. This is an iterative process as the largest releaserate may not give the worst-case consequences. Dependingon the complexity of the facility, experienced modelersshould identify such scenarios within 10-30 CFD dispersionsimulations. The larger of these gas clouds will thereafter beignited at different locations and exploded in order to evalu-ate the potential for DDT or high explosion pressures.Depending on possible release scenarios and the geometrylayout, the outcome of the realistic worst-case study willeither be comparable to the worst-case approach (if verylarge cloud sizes can be generated), or give lower conse-quences due to smaller cloud sizes or less ideal mixtures.

The third option will be a more comprehensive prob-abilistic study, which includes a ventilation study with 8–24different CFD simulations, a dispersion study with at least100–200 transient CFD simulations with a systematicvariation in the release parameters mentioned above,and finally an explosion study with approximately 100explosion simulations of idealized gas clouds of all sizesexpected to be generated in the dispersion study (varyingcloud location and ignition location). This approach is

24

similar to typical offshore probabilistic approach studies.The dispersion study coupled with ignition intensitymodels will produce probability of ignition for the differentgas cloud sizes, and the outcome of the complete study willinclude: risk of having a DDT; pressure-impulse or dragprobability of exceedance curves at various targets, pipework buildings, etc. This will provide a range of probabilityfor intolerable events.

In the FLACS CFD model, which is the most com-monly used consequence model for offshore explosionstudies, a new parameter, DPDX – the maximum normalizedspatial pressure gradient ahead of the flame – has been devel-oped to predict the potential for DDT, see example Figure 4.Based on validation studies (e.g., Middha et al., 2007) thepossibility of a detonation can exist when DPDX exceeds1, and should be expected once DPDX exceeds 10.Another condition necessary to achieve a DDT is that thenear homogeneous gas cloud that will support the detona-tion must have a relatively uncongested dimension of atleast 13 � 13 detonation cell sizes. While methane clouds(likely from LNG for dense gas behavior) will require anapproximately 4m thick layer, other relevant substanceslike propane and butane require a 1–2 m thick layer. The1–2 m thick layers are often present in scenarios that canpotentially detonate. This length scale criterion is thereforemostly relevant for methane and natural gas when it comesto DDT prediction for onshore facilities.

MITIGATIONMitigation may be evaluated if intolerable risk is identifiedor if improvements are sought to render the risk as lowas reasonably practicable. Mitigation measures include


Figure 4. Illustration of DDT criterion DPDX, FLACS simulation of Fraunhofer-ICT hydrogen lane tests (Middha et al., 2007), upper

plot shows pressures, middle plot flame position, while lower plot shows predicted DDT propensity DPDX as function of location.

reducing: the possibility for large gas clouds; the likelihoodDDTs; and also high pressure levels inside and outside thefacility.

If the potential for DDT is identified as a major chal-lenge in the risk study, there may be a range of differentapproaches to solve this. Quite often the most powerfulapproach will be to reduce the likelihood of the significantgas clouds that could potentially experience DDT or highpressures, while another approach would be to modify thedesign so that explosions are less likely too accelerate todetonation. Some possible measures are discussed below.

At least two explosion accidents mentioned in theintroduction seem to have experienced DDT’s, likelycaused by flames accelerating through trees. Vegetationinside and near facilities should clearly be considered in arisk study, and it may be recommended to remove treesand bushes near a facility, keep only tall trees with singletrunks and congestion above 4–5 m, or use 3–4 m tallvapour fences to prevent gas from entering arrays of treesand bushes. FLACS simulations of the Buncefield explosionshowed that the effect of these three mitigation measures allreduced the simulated explosion pressures by 1-2 orders ofmagnitude (Davis et al., 2010).

Inside the congested region of the plant, methodslike soft barriers may be utilized. A soft barrier is a set ofgas tight curtains, which will control gas migration, yetyield and vent in the presence of an explosion. If designedproperly, a system with soft barriers may limit the cloudsize inside a congested unit, and thus reduce the propensityfor DDT and high pressures. However, as the soft barrieralso will reduce ventilation, there is a potential for somewhatstronger explosions from smaller releases, and a proper studyshould be performed to evaluate this method. BP is using softbarriers on some of their offshore oil and gas platforms.

25

Many layout changes can be considered in the plant toreduce explosion risk, such as:

. Evaluate wall removal or change major decks from solidto grated or vice versa. Less confinement will usuallyreduce risk by influencing the cloud size distributionand limit flame acceleration, however, in many casesthe opposite effect is observed;

. Evaluate spacing between units, as well as optimizetheir size and shape LxWxH;

. Flashing liquid releases (e.g., LPG) may represent amajor concern as release rates and inventories can belarge. A strong cabinet can be built around the mainsources of release (e.g., tanks) so that any flashingrelease will impinge and lose its momentum beforeleaving the cabinet as a very fuel rich mixture fromthe bottom (with additional liquid particle collectionsystems). This could in most cases prevent the gener-ation of massive vapour clouds filling a large facilityat very reactive concentrations;

. Another potentially risk-reducing design for densevapours is to elevate units with congestion 4–5 mabove ground. This will leave an open area below theunits, with significantly better ventilation. As most flam-mable gas/aerosol clouds will be quite dense, the cloudwill fall down after losing momentum, and be efficientlydiluted and transported away below the congestedvolumes. If an explosion occurs in the elevated units,there will be additional pressure venting downwards tolimit flame acceleration. Such ideas have beenimplemented on gas processing plants.

Water deluge activated by threshold gas detectionlimits, are used to mitigate explosion consequences onmany offshore oil and gas platforms. In fact, ISO 13702


(1999) requires the potential effect of water deluge onexplosion mitigation to be evaluated as part of the riskassessment for offshore platforms. Large scale experiments(Al-Hassan et al., 1998) demonstrated that water mitigationwas even more efficient for onshore type facilities than fortypical offshore modules, due to less confinement. Explosionpressures were reduced from more than 10 barg to less than0.5 barg. Tests also demonstrated that water curtains atregular intervals could potentially control flame speeds. Foronshore facilities, which could have access to sufficientamounts of water (�10 l/sqm/min), it would be recom-mended to evaluate the potential benefits from water deluge.

To summarize this section, there are numerous waysto mitigate an explosion, which include influencing thegas cloud build-up, the explosion severity or both. Often-times the combination of two or more mitigation measuresis better than the sum of their individual effects. Oneexample of such an occurrence will be the combined

26

effect of reducing confinement and applying water miti-gation. Mitigation will typically have a negative effectfor some scenarios and positive effect for other. In orderto assess total combined effects of mitigation, a comprehen-sive study will be recommended, preferably with some kindof design-based or probabilistic approach.

While comprehensive studies may be consideredexpensive for facilities that have been relying on traditionalblast curve approaches, their cost will be low comparedto any design modification. If design change for mitigationis based on inappropriate consequence studies, the modifi-cation may have limited or no risk reducing effect, andthere may likely be much more cost effective ways to limitthe risk. More comprehensive safety studies throughout theonshore industry will also stimulate innovation, since itwill be possible to evaluate the effects of design changes.

If major accidents can be prevented, this shouldalso be of significant value to the facility. Buncefield,

Figure 5. Example of probabilistic risk study on an onshore facility (Hoorelbeke, 2006), upper picture shows simulation model of

propylene unit while lower picture shows cumulative frequency of gas cloud size based on CFD ventilation and dispersion study.


BP Texas City and Deepwater Horizon have remindedus about the potential losses when experiencing a majoraccident.

DISCUSSION AND CONCLUSIONExplosion risk assessment approaches for onshore facilitieshave been discussed. Based on several serious recentaccidents, where DDTs were concluded to have occurredin two of the incidents, we raise the question whether thecurrent simplified risk assessment approach for onshorefacilities is adequate. In most cases, risk assessmentsnever consider the possibility of DDT in their risk studies.

Current approaches for onshore facilities, which usequite simplified consequence tools that do not addresshow design and layout changes can reduce explosion risk,do not stimulate innovation towards safer designs and con-cepts. In that respect, the offshore oil and gas industry gen-erally has a very different philosophy using more functionalor risk based design and requirements for validated toolsand methods.

The Seveso-II directive requires member states toensure that “operator is obliged to take all measures necess-ary to prevent major accidents and limit their conse-quences”. It is quite difficult to achieve this goal usingsimplified approaches.

The Netherlands recently standardized the onshorerisk assessment approach, so that everybody has to use thesame package of integral tools for dispersion and blast-curves for explosion, neither of which can account fordetails in the geometry (i.e., plant layout and design).In France, there is currently a discussion whether to ruleout the use of 3D CFD tools for onshore blast and disper-sion studies, with the reasoning the French authorities(and advising research organizations) think these arenon-conservative and have a too high degree of user-dependency. At the same time, some major oil companiesseeing the benefits with the more comprehensive risk assess-ment approaches used offshore, are gradually starting to usethese approaches for their onshore facilities, see Figure 5(Hoorelbeke, 2006).

REFERENCESAl-Hassan, T. Johnson, D.M. (1998), “Gas explosions in large

scale offshore module geometries: Overpressures, mitigation

and repeatability, presented at OMAE-98, Lisbon, Portugal.

API-RP 752 (2009), Management of Hazards Associated

with Location of Process Plant Permanent Buildings, the

American Petroleum Institute.

BMIIB (2009), Buncefield Major Incident Investigation Board

(2009), the Buncefield Final Report.

27

Davis, S.G., Hinze, P., Hansen, O.R. and van Wingerden,

C. (2010), 2005 Buncefield Vapour Cloud Explosion:

Unraveling the Mystery of the Blast, presented at InterFlam,

Nottingham, UK.

Desbordes, D. (1995), Critical Initiation Conditions for

Gaseous Diverging Spherical Detonations, Journal de Physi-

que IV, Colloque C4, supplement to Journal de Physique III,

Volume 5, May 1995, pp. 155–162.

Hansen, O.R., Hinze, P., Engel, D. and Davis, S. (2010), Using

computational fluid dynamics (CFD) for blast wave predic-

tions, Journal of Loss Prevention in the Process Industries,

doi:10.1016/j.jlp.2010.07.005.

Hansen, O.R. & Wilkins, B. (2004), An experimental study on

oil mist explosions. International European Safety Manage-

ment Group (ESMG) Symposium, 16–18 March 2004,

ISBN 3-9807567-3-4, Nurnberg, Germany.

Hoorelbeke, P., Izatt, C., Bakke, J.R., Renoult, J. & Brewerton,

R.W. (2006), Vapour cloud explosion analysis of onshore

petrochemical facilities. ASSE-MEC-0306-38.

HSE (1975), ‘The Flixborough Disaster: Report of the Court of

Inquiry’, HMSO, ISBN 0113610750, Health and Safety

Executive, 1975.

ISO 13702 (1999), Petroleum and natural gas industries –

Control and mitigation of fires and explosions on offshore

production installations – Requirements and guidelines,

the International Organization for Standardization.

ISO 19901-3 (2010), Petroleum and natural gas industries –

Specific requirements for offshore structures – Part 3:

Topsides structure, the International Organization for

Standardization.

Middha, P., Hansen, O.R. & Schneider, H. (2007), Deflagration

to detonation transition (DDT) in jet ignited hydrogen-air

mixtures: large scale experiments and FLACS CFD predic-

tions. Twelfth International Loss Prevention Symposium,

22–24 May 2007, Edinburgh, UK, IChemE Symposium

Series 153.

Ontario Fire Marshal (2010), Sunrise Propane explosion

investigation report, investigation no: 212-007-2008.

Pierorazio, A.J., Thomas, J.K., Baker, Q.A. and Ketchum, D.E.

(2005), An update to the Baker-Strehlow-Tang vapour cloud

explosion prediction methodology flame speed table, Process

Safety Progress, Volume 24, Issue 1, pp. 59–65.

Puttock, J.S. (1995), Fuel gas explosion guidelines, the

congestion assessment method. In: Second European conf.

on major hazards on- and offshore, Manchester, UK, 24–26

September.

Seveso-II Directive 96/82/EC on the control of major-accident

hazards involving dangerous substances, The European

Commission.

van den Berg, A.C. (1985), “The Multi-Energy Method: A

Framework for Vapour Cloud Explosion Blast Prediction,”

Journal of Hazardous Materials, Vol. 12, pp. 1–10.

Use of CFD in onshore facility explosion siting studies

Documents